Abstract The complex nonlinear characteristics contained in the length-of-day variations parameter sequence seriously affects prediction accuracy.In order to explore whether the introduction of atmospheric angular momentum sequence can help improve the prediction accuracy, this paper proposes a Prophet fitting extrapolation joint vector autoregression residual compensation combined model to predict length-of-day variations. The sequence between 2008 and 2020 is selected for predicting experiments. At the same time, two schemes of Prophet-AR and traditional LS-AR, which ignore the atmospheric angular momentum sequence, are designed for comparison. The results show that the prediction accuracy of the three schemes decreases successively, which shows that Prophet algorithm can better fit the nonlinear signal than the LS algorithm to reduce the prediction error of the combined model. It also shows that the introduction of the atmospheric angular momentum sequence can effectively improve the prediction accuracy when the prediction model is consistent. It is comprehensively shown that the Prophet-VAR combined prediction model, which takes into account the atmosphere angular momentum, can be applied to high-precision prediction of length-of-day variations.
